Main idea

To find out what is missing from the description of a product, you enumerate it from several independent angles and measure, for each one, the share that corresponds to a card: the code modules, the database tables, the events emitted, the screens, the resources exposed by the API, the second application, the words customers use. Each angle brings back what no other one sees — the events give you the verbs of the business, the API gives you the objects scattered across several modules, the second application gives you the ones the first application's division ignores.

Taken on its own, a rate says almost nothing: it depends on an approximate recognition of names and on a mapping list kept by hand. The intelligence lies elsewhere. An object present in one angle and absent from the others is either a business object you missed, or plumbing — and the question is settled in one line, explicitly.

The number therefore serves to compare the angles with each other. A markedly lower rate on the second application doesn't mean "all that work is left": it means "this application handles things the rest of the product ignores, go and look over there".

Why it matters

This shifts what you expect from a coverage measurement. It proves nothing — it points to where to go and look — and that is exactly what makes it useful: nobody tries to push up a number that isn't a grade.

It also gives you a reason to multiply the angles rather than perfect a single one: the value comes from their independence, not from their precision.

Nuances and limits

The independence of the angles is an assumption. Two angles derived from the same structure — the tables and an export of those tables — never contradict each other and give a false impression of confirmation.

And the measurement itself stays fragile: the matching is done on names that resemble each other about half the time, topped up by a mapping list that is a judgement — a forgotten line makes a covered object show up as missing, an indulgent line inflates the score.

Open questions

• How do you verify the real independence of two angles before treating their agreement as a confirmation?